Version: 2.1.8
Elephant Bird is Twitter's open source library of LZO, Thrift, and/or Protocol Buffer-related Hadoop InputFormats, OutputFormats, Writables, Pig LoadFuncs, Hive SerDe, HBase miscellanea, etc. The majority of these are in production at Twitter running over data every day.
Apache licensed.
- git clone
- ant
- check out javadoc, etc.
- Play with the examples: ant examples
Note: For any of the LZO-based code, make sure that the native LZO libraries are on your java.library.path. Generally this is done by setting JAVA_LIBRARY_PATH in pig-env.sh or hadoop-env.sh. You can also add lines like
PIG_OPTS=-Djava.library.path=/path/to/my/libgplcompression/dir
to pig-env.sh. See the instructions for Hadoop-LZO for more details.
There are a few simple examples that use the input formats. Note how the Protocol Buffer-based formats work, and also note that the examples build file uses the custom codegen stuff. See below for more about that.
NOTE: This is an experimental branch for working with Pig 0.8. It may not work. Caveat emptor.
Elephant Bird takes advantage of Github's raw interface and self-hosts a maven repository inside
the git repo itself. To use the maven repo, simply add https://raw.github.com/kevinweil/elephant-bird/master/repo as a maven repo in the system
you use to manage dependencies.
For example, with Ivy you would add the following resolver in ivysettings.xml:
<ibiblio name="elephant-bird-repo" m2compatible="true"
root="https://raw.github.com/kevinweil/elephant-bird/master/repo"/>And include elephant-bird as a dependency in ivy.xml:
<dependency org="com.twitter" name="elephant-bird" rev="${elephant-bird.version}"/>- Protocol Buffers 2.3 (not compatible with 2.4+)
- Pig 0.8, 0.9 (not compatible with 0.7 and below)
- Hive 0.7 (with HIVE-1616)
- Thrift 0.5
- Mahout 0.6
Elephant Bird requires Protocol Buffer compiler version 2.3 at build time, as generated classes are used internally. Thrift compiler version 0.5.0 is required to generate classes used in tests. As these are native-code tools they must be installed on the build machine (java library dependencies are pulled from maven repositories during the build).
- JSON data
- Line-based data (TextInputFormat but for LZO; also available in deprecated 0.18 format)
- W3C logs
- Serialized Protocol Buffers and Thrift messages in one of three flavors:
- Block-based (also available in deprecated 0.18 format)
- Block-based, into generic bytes
- Line-based, base64 encoded
- Protocol Buffer and Thrift writables
- Serialized Protocol Buffers and Thrift messages in one of two flavors
- Block-based
- Line-based, base64 encoded
- LZO-only (patches to make this more general would be great)
- JSON data
- Regex-based loaders
- LzoPigStorage (just what it sounds like)
- W3C logs
- Serialized Protocol Buffers
- Block-based (dynamic or via codegen, see below)
- Line-based, base64 encoded (dynamic or via codegen, see below)
- In SequenceFiles, using ProtobufWritableConverter
- Serialized Thrift
- Block-based (dynamic)
- Line-based, base64 encoded (dynamic)
- In SequenceFiles, using ThriftWritableConverter
- SequenceFile loader
- Includes converter interface for turning Tuples into Writables and vice versa
- Provides implementations to convert generic Writables, Thrift, Protobufs, and other specialized classes, such as Apache Mahout's VectorWritable.
- LZO compression via LzoPigStorage
- Serialized Protobufs and Thrift
- SequenceFile Storage (with converters, as above)
- Counters in Pig
- Protocol Buffer utilities
- Thrift utilities
- Conversions from Protocol Buffers and Thrift messages to Pig tuples
- Conversions from Thrift to Protocol Buffer's
DynamicMessage - Reading and writing block-based Protocol Buffer format (see
ProtobufBlockWriter)
We provide InputFormats, OutputFormats, Pig Load / Store functions, Hive SerDes,
and Writables for working with Thrift and Google Protocol Buffers.
We haven't written up the docs yet, but look at ProtobufMRExample.java, ThriftMRExample.java, people_phone_number_count.pig, people_phone_number_count_thrift.pig under examples directory for reflection-based dynamic usage.
We also provide utilities for generating Protobuf-specific Loaders, Input/Output Formats, etc, if for some reason you want to avoid
the dynamic bits.
Note: this is not strictly required for working with Protocol Buffers in Hadoop. We can do most of this dynamically. Some people like having specific classes, though, so this functionality is available since protobuf 2.3 makes it so easy to do.
In protobuf 2.3, Google introduced the notion of a Protocol Buffer plugin that
lets you hook in to their code generation elegantly, with all the parsed metadata available. We use this in
com.twitter.elephantbird.proto.HadoopProtoCodeGenerator to generate code for each Protocol Buffer. The
HadoopProtoCodeGenerator expects as a first argument a yml file consisting of keys and lists of classnames. For each
Protocol Buffer file read in (say from my_file.proto), it looks up the basename (my_file) in the yml file.
If a corresponding list exists, it expects each element is a classname of a class deriving from com.twitter.elephantbird.proto.ProtoCodeGenerator. These classes implement
a method to set the filename, and a method to set the generated code contents of the file. You can add your own by creating
such a derived class and including it in the list of classnames for the Protocol Buffer file key. That is, if you want
to apply the code generators in com.twitter.elephantbird.proto.codegen.ProtobufWritableGenerator and
com.twitter.elephantbird.proto.codegen.LzoProtobufBytesToPigTupleGenerator to every protobuf in the
file my_file.proto, then your config file should have a section that looks like
my_file:
- com.twitter.elephantbird.proto.codegen.ProtobufWritableGenerator
- com.twitter.elephantbird.proto.codegen.LzoProtobufBytesToPigTupleGenerator
There are examples in the examples subdirectory showing how to integrate this code generation into a build, both for generating Java files pre-jar and for generating other types of files from Protocol Buffer definitions post-compile (there are examples that do this to generate Pig loaders for a set of Protocol Buffers).
Reading and writing Hadoop SequenceFiles with Pig is supported via classes SequenceFileLoader and SequenceFileStorage. These classes make use of a WritableConverter interface, allowing pluggable conversion of key and value instances to and from Pig data types.
Here's a short example: Suppose you have SequenceFile<Text, LongWritable> data
sitting beneath path input. We can load that data with the following Pig
script:
REGISTER '/path/to/elephant-bird.jar';
%declare SEQFILE_LOADER 'com.twitter.elephantbird.pig.load.SequenceFileLoader';
%declare TEXT_CONVERTER 'com.twitter.elephantbird.pig.util.TextConverter';
%declare LONG_CONVERTER 'com.twitter.elephantbird.pig.util.LongWritableConverter';
pairs = LOAD 'input' USING $SEQFILE_LOADER (
'-c $TEXT_CONVERTER', '-c $LONG_CONVERTER'
) AS (key: chararray, value: long);
To store {key: chararray, value: long} data as SequenceFile<Text, LongWritable>, the following may be used:
%declare SEQFILE_STORAGE 'com.twitter.elephantbird.pig.store.SequenceFileStorage';
STORE pairs INTO 'output' USING $SEQFILE_STORAGE (
'-c $TEXT_CONVERTER', '-c $LONG_CONVERTER'
);
For details, please see Javadocs in the following classes:
- SequenceFileLoader
- SequenceFileStorage
- WritableConverter
- GenericWritableConverter
- AbstractWritableConverter
Bug fixes, features, and documentation improvements are welcome! Please fork the project and send us a pull request on github.
Each new release since 2.1.3 has a tag. The latest version on master is what we are actively running on Twitter's hadoop clusters daily, over hundreds of terabytes of data.
Major contributors are listed below. Lots of others have helped too, thanks to all of them! See git logs for credits.
- Kevin Weil (@kevinweil)
- Dmitriy Ryaboy (@squarecog)
- Raghu Angadi (@raghuangadi)
- Andy Schlaikjer (@sagemintblue)
- Travis Crawford (@tc)